1

Word Meaning and Concept Expressed

Robyn Carston
Linguistics, University College London, and
Centre for the Study of Mind in Nature, University of Oslo

Abstract:
The concept expressed by the use of a word in a context often diverges from its
lexically encoded context-independent meaning: it may be more specific or more
general (or a combination of both) than the lexical meaning. Grasping the intended
concept involves a pragmatic process of relevance-driven adjustment or modulation
of the lexical meaning in interaction with the rest of the utterance and with contextual
information. The issue addressed here is the nature of the input to the pragmatic
process of meaning adjustment, that is, the nature of the standing (encoded) meaning
of the word type. The widespread assumption that lexical meaning is conceptual,
hence directly expressible, is challenged and a case made for the merits of an account
of word type meaning in non-conceptual terms.

1. What is a word meaning?

We use sentences to express/communicate thoughts (truth-conditional contents) and we

use words to express/communicate concepts, which are constituents of thoughts (hence
contribute to truth-conditional contents). It is now quite widely accepted that the meaning
(or semantic content) that a word is used to express or communicate on an occasion of
utterance is often distinct from the meaning it has as an expression type in a language
system (that is, its standing or encoded meaning). This view is shared by ‘contextualist’
philosophers of language, by some linguists, and by pragmaticists working within the
cognitive-scientific framework of relevance theory. The main aim of this paper is to


I thank Mark Textor, Ruth Kempson and Timothy Pritchard for very helpful discussions. I am also grateful
to two anonymous referees for Linguistic Review and, in particular, to the editor of this special issue, Stavros
Assimakopoulos, for his sound advice and constant encouragement. This work is supported by an AHRC
Research Grant, AH/I000216/1, awarded to the project: Word Meaning: What it is and what it is not.
 2

consider the nature of the context-free word meaning which is the starting point for the
pragmatic processes that deliver the occasion-specific meaning (the concept meant or
communicatively intended by the speaker).
I am confining my attention here to what are often called ‘open class’ words, that
is, words with an apparently descriptive meaning, such as nouns, verbs, adjectives, and
most adverbs, leaving aside ‘closed class’ words, such as indexicals, determiners, function
words, and connectives. With the domain so restricted, the question is: ‘what is a word
meaning?’ Word type meanings might be concepts, hence contentful entities that can be
constituents of thoughts. If that is the right answer, then standing word meanings are the
same sort of thing as that which we use them to express/communicate (concepts,
semantically evaluable entities) and there is no reason why speakers would not, at least
sometimes, use words to express the very meaning they encode. This view is held by some
psychologists (e.g. Murphy 2002), by some philosophers (e.g. Fodor 1998) and by
relevance theorists (e.g. Sperber and Wilson 1995, 1998). However, there is an equally
widely held view that word meanings are ‘underspecified’; that is, that they cannot
contribute directly, without modification or transformation of some sort, to the
thoughts/propositions that utterances in which they occur are used to express. There are
various possibilities for what these underspecified entities might be: a special kind of
‘lexical’ concept, a pro-concept, a schema or procedure or set of constraints on the kind of
contentful concept they can be used to express/communicate. Any of these could qualify
as a word meaning, as a specification of a word’s ‘semantic potential’. However, a more
radical position is that words (lexical forms) do no encode concepts or abstract schemas or
constraints, but are associated with something else altogether, something that does not
qualify as a meaning of the expression type. Two apparently rather different possibilities
that have been suggested are (a) collections of memory traces or exemplars of previous
uses (tokenings) and (b) bundles of contingent encyclopaedic information about the things
in the world the word is used to refer to.
These four positions (ordinary concepts, lexical concepts, abstract schemas or
constraints, previous uses or encyclopaedic information) can all be seen as falling within a
broadly contextualist view of word meaning and they correlate roughly with the four
contextualist positions set out by Recanati (2004). The most conservative of these is the
‘strong optionality’ position (quasi-contextualism), according to which a word’s meaning
may contribute directly (unmodulated) to truth-conditional content or may be
 3

pragmatically modulated/adjusted. The second is the ‘pragmatic composition’ view,

according to which a word meaning (lexical concept) could be an expressed sense, but the
process of composing it together with the other words in the utterance forces its pragmatic
adjustment. On the third, more radical, view which Recanati calls the ‘wrong format’
position, word meanings cannot enter directly into thought, but must be transformed into
the right (conceptual/contentful) format, presumably by some pragmatic interpretive
process or other. Here, there is a distinction between the ‘semantic structures’ of linguistic
expressions and the conceptual structures of thought. If word meanings are abstract
schemas or sets of constraint or rules for use, then they are in the ‘wrong format’ to be
semantically contentful or to be constituents of thoughts. Finally, there is the most extreme
position, that of meaning eliminativism, according to which words (qua types) do not have
meanings at all; only tokens (specific utterings) of words have meanings.
It is instantiations of these latter two broad positions that I want to consider as
possible candidates for what word types bring to the pragmatic processes of utterance
understanding. I will present a range of considerations that point in the direction of there
being something less than a fully conceptual meaning for words, thus ruling out the first
two positions. I have no knock-down arguments and there are some obvious pitfalls to
taking on this non-conceptual view. However, I do think it merits serious consideration,
especially when placed in the context of relevance-theoretic lexical pragmatics, which
offers an explanatory account of how the concept a speaker expresses with a word can
vary across occasions of use.1

2. Relevance theory pragmatics

The view that sentence meaning (logical form) is seldom, if ever, fully propositional has
been a basic tenet of relevance theory (RT) since its inception. The claim is that it is a
propositional template or radical, which must be pragmatically completed and elaborated
on each occasion of utterance in order to derive the propositional content meant

1
An anonymous referee has suggested that I should make it explicit early on that, in this paper, I am
treating all (open class) words as having the same kind of meaning and not discriminating between those
words that are (semantically) polysemous and those that are not. In fact, I am far from sure that any such
distinction should be made: my working hypothesis is that (open class) words quite generally are susceptible
to pragmatic adjustment in context and that some, a minority, of these (initially ad hoc) derived senses
become routinized or conventionalized to varying degrees, due to repeated use. What we describe as
‘polysemous words’ are those whose several senses have crossed some threshold of frequency or
conventionality.
 4

(communicatively intended) by the speaker (Carston 2002). Recently, the focus has been
more on word meaning and a detailed RT account of lexical pragmatics has been
developed (Wilson and Carston 2007). Consider the examples in (1), a set of utterances
about Boris, who, let us suppose, is a man in his forties and has been married for many
years:2

(1) a. Boris is a man.

Encoded meaning: B IS A MAN
Explicature: B IS A MAN*
b. Boris is a child. CHILD*

c. Boris is a bachelor. BACHELOR*

d. Boris is a chameleon. CHAMELEON*

In (1a), the linguistically encoded content is a trivial truth, hence uninformative,

insufficiently relevant. The addressee’s process of trying to derive contextual implications
from the utterance, in accordance with his expectation of relevance, leads to the encoded
concept MAN being narrowed down so as to encompass just men of some kind. Depending
on the specifics of the context, it could be narrowed down to ‘typical man’ or ‘ideal man’
and, of course, what constitutes a typical man or an ideal man will itself vary from context
to context. The outcome of this process is an occasion-specific sense (or ‘ad hoc’ concept)
MAN*, which picks out a proper subset of the set of individuals that fall under the original
encoded concept MAN.
In (1b), we have the opposite phenomenon: the encoded concept CHILD is adjusted
so as to mean roughly ‘person who behaves in certain childish (or child-like) ways’, and
the result is a concept CHILD* which is broader than the lexically encoded concept - it
includes actual children and some adults. Then, if we take (1c) as an utterance by Boris’s
wife, who has long endured his affairs with other women and general lack of commitment,
this is, arguably, both a broadening of the lexical concept BACHELOR (it includes married
men who behave in certain ways) and a narrowing (it excludes bachelors who don’t
behave in this stereotypic way). Finally, (1d) is a typical metaphorical use, which in
standard RT is taken to be a radical kind of broadening, so the concept communicated,

2
The ‘*’ on the concept expressed by the speaker of these utterances is just a notational device to indicate
that it is distinct from the lexically encoded concept and was derived by a pragmatic process.
 5

CHAMELEON*, roughly paraphraseable as ‘individual that can alter appearance or

behaviour so as to fit in with current surroundings or circumstances’, includes actual
chameleons (a kind of lizard) and certain human beings (and any other creatures with the
property at issue).3
These pragmatically-derived (ad hoc) concepts are components of the speaker’s
explicature, or what Recanati (2004) speaks of as the pragmatically enriched ‘what is
said’, the intuitive truth-conditional content of the utterance. They are recovered by the
addressee in the process of finding an interpretation that meets his context-specific
expectations of relevance (licensed by the general presumption of ‘optimal relevance’
carried by all utterances).4 According to the account, there is a single process of lexical
concept adjustment (or meaning modulation), which can have any of several outcomes: it
can result in a concept whose denotation is narrower or broader (or both) than that of the
lexical concept.5 The process is a kind of ‘free’ pragmatic enrichment; that is, unlike the
pragmatic fixing of a value for an indexical, it is not linguistically mandated or controlled.

Now, let’s consider what it is that words, qua expression types in an individual’s language
system, bring to the process of interpreting particular utterances; that is, what the
lexically-based input to pragmatics is, according to the relevance-theoretic approach
(Sperber and Wilson 1998; Wilson and Carston 2007). In effect, there are two parts to this,
a semantic part, the encoded meaning of the word, and a contingent, non-semantic part,
consisting of encyclopaedic information associated with the encoded meaning. Let’s look
at these in turn.

3
An anonymous referee has objected to this account of the metaphorical use of the word ‘chameleon’,
maintaining that the concept that it expresses pertains just to a certain kind of human being (so would not
include actual chameleons in its denotation). I am broadly sympathetic to this position and am currently
developing an account of metaphorical language use according to which it requires both broadening and
narrowing, so that the denotation of the ad hoc concept derived may merely overlap with that of the lexical
concept or may be entirely disjoint from it (see Carston and Wearing [2011] for a preliminary formulation of
this idea). However, the position outlined above is the established ‘loose use’ account of orthodox relevance
theory (see, for instance, Sperber and Wilson [1995]).
4
For a recent outline of the principles and mechanisms of relevance theory, see Carston 2012.
5
Several people have raised objections to the construal of concept broadening and narrowing in
denotational terms (Kempson [p.c.] and Textor and Allott [forthcoming]). I see this externalist semantic way
of characterising these notions as a first shot, made in the interests of presentational clarity, but as only a part
(and a part that may need to be modified) of a much more extensive account of ad hoc concepts whose
internalist mental representational details must ultimately be given.
 6

Most words are taken to encode concepts, specifically atomic concepts. There is a
simple mapping from lexical form, e.g. /kæt/, to mental concept, e.g. CAT, the concept is
unstructured and the lexical entry does not specify any further information about its
content or semantic behaviour. In short, the position is essentially the same as Jerry
Fodor’s ‘disquotational lexicon’: the word ‘house’ means HOUSE, ‘miserable’ means
MISERABLE, ‘keep’ means KEEP, etc. (Fodor 1998; Fodor and Lepore 1998), and most of
the details of Fodor’s view of concepts are preserved: Concepts are mental particulars
(they function as mental causes and effects); they categorise the world so applications of
concepts are susceptible of ‘semantic evaluation’ (as true/false; correct/incorrect); concept
content is wholly referential (not constituted by inferential relations);6 concepts are
constituents of thoughts and of each other (phrasal concepts); thoughts (and phrasal
concepts) inherit their content from the contents of their constituent concepts. In a sense,
then, words (lexical forms) don’t have a semantics – they merely inherit a semantics from
the concepts they encode and the ‘real’ semantic story is about the concept-property
(mind-world) relation, hence what it is to possess a concept and how we come to possess
them, that is, what the mechanisms are through which a symbol in the head locks onto a
property in the external world.
Associated with concepts (whether lexicalised or not) are collections of
encyclopaedic information, including general knowledge and individual beliefs about the
things they denote, cultural knowledge, including stereotypes, which the individual may or
may not endorse, imagistic representations, and perhaps also episodic memories. (Much
more needs to be said about how this information is organised and tagged so that general
knowledge, stereotypes, individual memories, etc. are kept distinct from each other.) As
Fodor (2008) puts it, concepts can be thought of as names or labels for files containing
such collections of contingent information, so that: ‘In effect, according to this story, we
think in file names;’ (Fodor 2008: 95).
Encyclopaedic information plays a key role in the relevance-theoretic process of
lexical concept adjustment. When a lexical concept is accessed via the usual linguistic
decoding process, the encyclopaedic information associated with it is activated. Some
elements of it are more highly activated than others (since there are multiple sources of

6
Here RT does depart from Fodor in assuming that many concepts come with a logical entry, a set of
inference rules that capture certain necessary conditions on a concept’s content. I can and will ignore this
complication for the purposes of this paper.
 7

spreading activation, including other concepts encoded in the utterance and conceptual
representations derived from the wider discourse or situation of utterance). The most
highly activated items of conceptually represented information are accessed and deployed
as contextual assumptions in deriving contextual implications, which form an initial
interpretive hypothesis about the utterance. Then, via a mechanism of mutual parallel
adjustment of explicit utterance content (explicature), contextual assumptions and
contextual implications, concepts in the decoded logical form are adjusted by backwards
inference, so that only implications that are ultimately grounded in the explicature are
confirmed. The overall interpretation is accepted provided it meets the addressee’s
expectation of relevance. Consider again example (1b) ‘Boris is a child’. Depending on
the wider discourse situation, contextual implications such as Boris is sweet and innocent,
untouched by life experience, may be inferred, based on assumptions accessed from the
encyclopaedic entry for CHILD, which, by backwards inference, lead to a particular ad hoc
concept CHILD*. In another utterance situation, different items of encyclopaedic
information about children might be more highly activated making most accessible such
implications as that Boris doesn’t earn his keep, expects others to look after him, is
irresponsible, etc., resulting in a distinct ad hoc concept CHILD** in the explicature. And
there are other – perhaps indefinitely many – possibilities.
There’s no arguing with the existence and importance of this kind of general world
knowledge associated with words and concepts, nor with the claim that it is contingent,
that is, it is extrinsic to the concept’s content and plays no part in its individuating
conditions. The part of the story that one could take issue with is the linguistic semantic
part; that is, the claim that word type meanings are concepts. In the next section, I’ll set
out some considerations that make an alternative non-conceptual account worth serious
investigation.

3. Considerations in favour of a schematic (underspecified) lexical meaning

I take it that concepts are, first and foremost, constituents of thoughts. This is in
accordance with Fodor’s view of them as, in effect, words of Mentalese (the language of
thought) and with the views of most people working within cognitive science, including
relevance theorists. So the first point I want to consider here concerns the role of the
hypothesised encoded word meanings (lexicalised concepts) as components of thought.
 8

Take the concept HAPPY, for instance, (allegedly) encoded by the word ‘happy’; this
provides communicative access to a wide range of other more specific concepts, including
one for a steady state of low-key well-being (HAPPY*), another for a momentary
experience of intense joy (HAPPY**), another for the sense of satisfaction that
accompanies a successful transaction or completion of a job (HAPPY***), and so on. All
of these specific concepts are components of thoughts we might have. The question then
is: Does the very general lexically-encoded concept HAPPY occur as a component of
thought and, if so, what sort of thought is this?
Similar questions arise for the (alleged) concept OPEN encoded by the verb ‘open’.
In an interesting discussion of pragmatic polysemy, Sperber and Wilson (1998: 197) say:
“A verb like ‘open’ acts as a pointer to indefinitely many notions or concepts ...” and they
mention cases for which the intended concept is jointly indicated by the verb and its direct
object, as in ‘open the door’, ‘open a letter’, ‘open a tin’, ‘open one’s mouth/one’s eyes’,
etc, and others which depend on broader non-linguistic context (e.g. when opening a door
might involve breaking it down with an axe, or opening one’s mouth might require
removal of stitches with which it has been fastened). However, they maintain that the
pragmatic adjustment process is optional because: “It may so happen that the intended
concept is the very one encoded by the word, which is therefore used in its strictly literal
sense” (Sperber and Wilson 1998: 196-97). So let’s consider the assumed lexical concept
OPEN and what it is to have a thought in which such a general concept features, as opposed
to any of the more specific concepts that we grasp in understanding ‘open one’s mouth’,
‘open the window’, etc. The question is whether there is any definite thought at all or
whether any thought about opening must contain one of the more specific concepts.
Consider the following, trying to construe them as thoughts containing a very general
lexical concept OPEN as a constituent:

(2) a. Whenever I open anything I feel anxious.

b. Everyone opens things sometimes.

As far as I can tell, the thoughts about ‘opening things’ that we take to be expressed by
these sentences are ones in which the ‘things’ at issue are construed as some sort of
coherent subcategory of all the things that one could talk about opening, so, e.g., it might
be the category of things that can contain stuff inside them, like boxes, envelopes, files,
 9

brief-cases, and cupboards. It seems unlikely to include opening one’s mouth or eyes (or
even opening curtains or windows, or gates), let alone opening discussions, lectures,
issues, minds, hearts or cans of worms. So while the ‘open’ concept that will figure in the
thoughts in (2) is indeed quite a broad one, it is still, I think, considerably narrower than
the supposed lexical concept that is allegedly encoded by the verb ‘open’ and provides the
basis for inferring all the more particular concepts of opening.
This first consideration has been presented in the form of a thought experiment and
it would, of course, be more satisfactory if it could be backed up by more empirically-
based evidence. This might be possible through behavioural experiments on human
categorization, which, if the thinking here is right, employs concepts of a finer-grain than
such very general lexically encoded concepts as OPEN and HAPPY.
The second consideration hinges on some well-aired problems with the
phenomenon of polysemy. Let’s take a fresh example, the word ‘stop’, as discussed by
Agustín Rayo, who notes that it can move across syntactic categories (verbal and
nominal):

You can stop writing; you can stop a burglar; you can stop a cheque; you can stop a
nail hole with plaster; you can use your fingers to stop the holes of a flute; you can
stop a poker into the fire; you can stop the tide by anchoring your boat. … You can
come to a stop; there can be a stop in your speech; you can include a stop in a
telegram; you can put a stop on a camera; you can pull out all the stops on an organ.
(Rayo 2011)

The language is full of such polysemous verbal-nominal words; consider, for instance,
‘fall’, ‘rest’, ‘cut’, ‘run’, ‘jump’, ‘skip’, ‘walk’, ‘start’, ‘end’, ‘turn’, ‘slip’, ‘pass’, ‘talk’,
sign’, ‘file’, and so on.7 As Rayo points out, this kind of grammatical and semantic
versatility has been tackled by computationally-minded linguists through the postulation
of an array of (sometimes quite complex) lexical-semantic rules (see, e.g., Pustejovsky
1995, Asher 2011).

7
On the account of the (‘exo-skeletal’) lexicon developed by the syntactian Hagit Borer, words as isolated
entities (or ‘listemes’, as she calls them) do not belong to specific syntactic categories but only acquire
syntactic status when they are used in particular syntactic structures (see Borer 2005). This view appears
likely to mesh better with the ideas being pursued here than the predominant view among syntacticians
which has much of the grammar projected from complex lexical entries (the ‘endo-skeletal’ view of the
lexicon).
 10

There is a major obstacle to any lexical rule approach which is that since the range
of concepts that a word can be used to convey is indefinite, commensurate with the
indefinite range of contexts human communicators can find themselves in, a line has to be
drawn between those that are to be accounted for by linguistic rules or conventions and
those that are to be left to the pragmatic ingenuity of language users. But, as Nunberg
pointed out long ago: “There is a substantial class of cases where we have no principled
grounds for deciding which of several uses is conventional, i.e. licensed entirely by
linguistics rules, [and which are derived]” (Nunberg 1979:154). The most problematic
manifestation of this issue is what he called ‘the non-uniqueness of semantic solutions’,
that is, cases for which there does not seem to be, among the related senses, one that is the
basic or central sense, from which the others are derived.
Consider the following senses of the word ‘window’:

(3) a. The bay windows are a beautiful feature of the house. [glass pane and
frame]
b. The cricket ball smashed my study window. [glass pane]
c. She crawled through the upstairs window and fell onto the floor.
[open space in wall]
d. The eyes are the window to the soul. [something one can see through]
e. We must seize this window of opportunity. [something of short duration]

Which (of the first three) of these is the lexical concept WINDOW from which the others are
pragmatically derived? None seems more intuitively basic than the others, nor the most
useful as a starting point from which the others could be pragmatically derived in context.
Cases of part-whole or metonymically related senses, which are surely among the most
ordinary, everyday and uncreative of our uses of language, seem to be highly problematic
in this regard. The noun ‘novel’, for example, discussed at length by Bosch (2007), can
have the following senses: a complex of ideas/thoughts (when the author is working on it),
a text (when it is completed), a publication (e.g. when we talk of an author’s most recent
novel), a physical object (e.g. when we talk of a suitcase full of novels), and certain
combinations, e.g. ‘Peter is reading the novel he found at the bus-stop’ (text and physical
object). Again, no particular one of these senses is obviously the encoded meaning or is
sufficiently all-encompassing to provide the basis for pragmatically inferring the other
 11

senses. As Bosch says: “If we want to maintain just one lexical entry for ‘novel’ it must
remain underspecified in many respects …” (2007: 59). And this point applies to a wide
range of other words (see Nunberg 1979, Bosch 2007, Lossius Falkum 2011). So, instead
of trying to force one of the multiple senses into the role of basic underived sense, perhaps
we should give up on assuming that there is one that plays this role, that is, give up on
assuming that there is an encoded lexical concept.8
A third consideration, one that has been pointed out by Sperber and Wilson (1998),
is that ‘words behave as if they don’t encode concepts’. First, they suggest that there are
many words that do not encode a full-fledged concept but what might be called a pro-
concept, giving as likely examples, ‘my’, ‘have’, ‘near’, ‘long’, and saying that “while
each of these examples may be contentious, the existence of the general category should
not be” (Sperber and Wilson 1998: 185). They don’t say anything more about what a pro-
concept is, but it’s clearly intended to be something less than a complete concept with a
referential content, something that requires that a semantic value is pragmatically inferred
in context, so it seems to be an indexical element of some kind. They then go on to say:
“… quite commonly, all words behave as if they encoded pro-concepts: that is, whether or
not a word encodes a full concept, the concept it is used to convey in a given utterance has
to be contextually worked out” (Sperber and Wilson 1998: 185). And, in later, more fully
developed work on lexical pragmatics, Wilson and Carston (2007: 231) maintain: “…
lexical narrowing and broadening (or a combination of the two) are the outcomes of a
single interpretive process which fine-tunes the interpretation of almost every word.” So
the question here is: if words quite generally behave as if they don’t encode concepts, why
maintain that they do encode concepts?
This is not just an issue for relevance theorists; it arises also in interestingly
parallel recent work by the formally-oriented linguist Peter Bosch (2007). He
distinguishes what he calls ‘lexical concepts’ from ‘contextual concepts’ and says: “Cases
of apparent variation in word sense require treatment at the conceptual level rather than a
lexical semantic solution.” He discusses a range of cases of nouns, including ‘novel’, as
discussed above, and of predicates, including ‘is working’:

8
Space limitations preclude discussion of the recent sophisticated and comprehensive account of word
meaning by Asher (2011), in which certain polysemous words are taken to encode a complex semantic type,
.
e.g. the lexical semantics of the word ‘book’ is of type [PHYSICAL OBJECT INFORMATION], which is one of
.
several complex types [α β].
 12

(4) a. Where is Fred? He’s working.

b. What is Fred up to today? He’s working.
c. How can Fred afford those expensive holidays? He’s working.
d. Fred is working.

As an answer to the different questions in (4a) - (4c), the utterance of ‘He’s working’ is
interpreted differently in each case: as giving information about Fred’s location (4a), about
Fred’s current activity (4b), and about Fred’s financial situation (4c). Then, in (4d), it is
interpreted differently depending on whether Fred is taken to be the name of our
building’s caretaker or of a prize race-horse or of my computer. As Bosch points out, these
different contextual concepts expressed by the word ‘work’, WORK1, WORK2, WORK3,
WORK4 …, are truth-conditionally relevant and licence different inferences. He maintains
that the lexical type meaning of ‘work’ and of many other words is ‘underspecified’, that
is, it must be developed at a conceptual or pragmatic level in order for its expressed
meaning to be realised. However, he persists in labelling it a ‘lexical concept’ and likens
his position to that of Fodor and Lepore’s (1998) ‘disquotational’ view of the lexicon
(lexical forms map directly to atomic concepts). Again, my question is: why insist that
words encode concepts?
The fourth and final consideration concerns semantic compositionality. As Fodor
has pointed out repeatedly, there are some fundamental properties of language and
thought, namely their systematicity and productivity, that can only be explained by the
semantic compositionality of these representational systems: there is a basic stock of
primitives (words/concepts) with stable semantic values and a recursive syntax such that
the semantic value (content) of any sentence/thought is a function of the semantic value of
the primitives and the way in which they are syntactically combined. With regard to the
‘compositionality’ of thoughts/concepts, Fodor (1998: 25-27) says: “Since it’s required to
explain productivity and systematicity, compositionality is, as one says, ‘not negotiable’.
An account of concept possession that is incompatible with the compositionality of
thought is, ipso facto, out of the running.” Thus, he has argued that an account of concept
content in terms of stereotypes or prototypes or inferential roles or partial definitions fails
because these entities do not meet the compositionality requirement.
 13

Now it might seem that this carries over point for point to public language systems
and so to word meanings. However, that is not the case. Natural language sentences are
simply not compositional in the required sense, that is, the propositions/thoughts they
express are not determined by word type meanings and syntax alone, as Fodor himself
occasionally acknowledges: “…a perfectly unelliptical, unmetaphorical, undeictic
sentence that is being used to express exactly the thought that it is conventionally used to
express, often doesn’t express the thought that it would if the sentence were
compositional. Either (the typical case) it vastly underdetermines the right thought; or the
thought it determines when compositionally construed isn’t, in fact, the one that it
conventionally expresses.” (Fodor 2001: 12; my emphasis [RC]). He concludes: “The
evidence suggests strongly that language is not compositional.” (Fodor 2001: 14).
As mentioned in Section 2, this has always been a central claim of relevance
theory and of the contextualist philosophers. What has not been so much noted is the
implication that it has for an account of word (type) meanings: the compositionality-based
arguments against the adequacy of prototypes, inferential roles, partial definitions, etc. as
the semantic content of concepts do not carry over to word meanings. Once you drop the
compositionality requirement on linguistic (sentence) meaning (while of course
maintaining it for thoughts), any requirement that a word (type) meaning contributes
content directly to truth conditions and so encodes a concept falls away. Words could
encode prototypes, or inferential relations, or schemas, or constraints, or any of a range of
other non-conceptual (non-contentful) possibilities, provided that the kind of component
required to preserve the compositionality of thought, that is, concepts (with a referential
semantics), can be delivered by pragmatic processes. In short, severing the relation
between word type meanings and concepts does not violate the non-negotiable
compositionality constraint.
I hope that collectively these considerations provide sufficient impetus to warrant
looking into non-conceptual characterisations of word type meaning.

4. Alternatives to concepts as word ‘meanings’

The four alternatives I’m going to mention briefly here are either instantiations of the
position Recanati (2004) calls the ‘wrong format’ view, i.e. linguistic ‘semantics’ does not
provide a truth-conditional component (a ‘content’), but still there is some stable, context-
 14

free meaning associated with the word type, or instantiations of the most radical
contextualist position, which he calls ‘meaning eliminativism’, i.e. there is nothing
resembling a stable word type meaning, but rather a collection of resources for concept-
making that the lexical form activates.
The first possibility to consider is that all these apparently descriptive words
‘behave as if they encoded pro-concepts’ because that is what they do in fact encode, that
is, they fall in with the general class of indexicals (which are subject to a pragmatic
process of saturation, of finding the appropriate semantic value in the context). This is not
an attractive solution, for several reasons. Indexicals constitute a specific small set of
words, whose context-sensitivity is entirely systematic, while practically the whole of the
descriptive vocabulary is modulated and in a non-systematic way. Indexicals come, in
effect, with a slot and an instruction on the kind of thing to plug into the slot and the
pragmatic process of slot-plugging is obligatory. The cases of context-dependence we are
considering here are quite different: the pragmatically-derived concepts for a particular
word can differ from each other in arbitrarily many ways and, even supposing we could
set out the full range of parameters of variance, it would not be obligatory (or possible) to
provide all of them with a semantic value on every occasion of use (Recanati 2004, Bosch
2007).
A second option is that a word type meaning is a ‘formal’ linguistic entity of some
sort. For instance, in a discussion of the lexical semantics for classes of verbs, Glanzberg
(2011) argues for a monadic conceptual root in a structural frame, along the following
lines:

(5) a. ‘X open Y’: [[X act ] cause [become [Y <open>]]]

b. ‘X hit Y’: [X act<hit> Y]

It is important to note here that the structural frames for particular classes of verbs (‘open’
and ‘hit’ belonging to distinct classes) are grammatically determined and the components
of the frames, act, cause, become, etc., are linguistic/grammatical elements, which are not
identical to the ordinary concepts ACT, CAUSE, BECOME, etc. Whether the conceptual roots,
‘<open>’, ‘<hit>’, etc., are semantic or syntactic elements is left open (Glanzberg 2011:
9). This is clearly, then, a case of ‘wrong format’, that is, of linguistic semantic
representations being a different kind of entity from conceptual (thought) representations,
 15

and the issue it raises, as for any other manifestation of this difference, concerns how we
make the move from the one to the other in communication.
A third option is that words encode something schematic: a template for concept
construction, a set of constraints, a rule for use, a sense-general meaning (or ‘archi-
sememe’ or ‘super-concept’), as variously discussed by Ruhl (1989), Moravcsik (1994),
and Atlas (1989). Ruhl, for instance, says: “… lexical meaning must be highly abstract
(though still specific to a particular language), and thus highly formal, … remote from all
ambient contingencies” (1989: ix). It may be that Bosch’s (2007) ‘lexical concepts’, which
do not determine an expressible content, also fall in here. No doubt, I am grouping
together a disparate range of quite different manifestations of the general idea of stable
word ‘meanings’ that underspecify but constrain the kinds of contents we can
communicate when we use them. I cannot explore these different notions here, but note
merely that something along these lines seems to be quite widely favoured.9
Under the ‘wrong format’ position, Recanati (2004) talks, on the one hand, of
meanings that are too abstract and schematic, and, on the other, of meanings that are too
rich, incorporating a host of ‘semantic features’, which have to be whittled down
(cancelled) by contextual/pragmatic considerations on any occasion of use in order to
recover the particular sense expressed/communicated. Cohen (1993) takes a view of this
latter sort, so that, for instance, the noun ‘rose’ has the features [plant, flower, petals,
thorns on stem, beautiful, fragrant, highly valued …] which are ordered in terms of
centrality or prominence. When, for instance, the word is used metaphorically, or
otherwise loosely, some of these features are cancelled. Evidently, Cohen includes in his
set of ‘semantic features’ some components of what seem to be contingent encyclopaedic
information (about roses, for instance), in an attempt to provide all the ingredients needed
for different uses. Given that we have a vast store of encyclopaedic information about real
world entities (e.g. roses) and that innovative uses of words (expressing genuinely new ad
hoc concepts) are always possible, it seems that, in its bid to be comprehensive, this
approach must ultimately collapse into a variety of what Recanati calls the ‘meaning
eliminativist’ position.

9
Another account within this broad category, one that warrants detailed discussion, is that of Ruth
Kempson, Eleni Gregoromichelaki and Christine Howes (2011), who characterise word type meanings as
‘lexical actions’ or procedures, which together with the instructions provided by the syntax of a language
constitute a set of mechanisms enabling the construction of representations of content by the interpreter in
the process of utterance comprehension.
 16

A clearly eliminativist position is the recent ‘grab-bag’ model of Rayo:

With each expression of the basic lexicon, the subject associates a ‘grab-bag’ of
mental items: memories, mental images, pieces of encyclopaedic information,
pieces of anecdotal information, mental maps and so forth. With the expression
‘blue’, for example, a subject might associate two or three particular shades of
blue, the information that the sky is blue, the information that my bicycle is blue,
a memory of a blue sweater, and so forth. Different speakers might associate
different grab-bags with the same lexical item. (Rayo 2011)

This looks very much like the kind of information associated with a word that is given as
an encyclopaedic entry or material in a mental file in accounts that assume words encode
concepts, as in relevance theory or Fodor (2008). It surely doesn’t qualify as the linguistic
meaning or semantics of a word type – it is totally non-linguistic and largely contingent.10
An appealing aspect of this approach is that it seems to provide an immediate and
simple solution to the polysemy/metonymy problem. Referring to his example of the
polysemous verb-noun ‘stop’, mentioned above, Rayo says: “One can place a few key
items in one’s grab-bag for ‘stop’ – for instance, representations that bring to mind
interfering, preventing, obstructing, closing – and let common sense and sensitivity to
context take care of the rest.” And there is no need for different grab-bags for different
grammatical categories: “a mental image that evokes obstruction, for example, can be used
to render salient the action of closing a valve when interpreting … ‘she stopped the flow
of oxygen’ and to render salient a knob on a pipe organ when interpreting … ‘she moved
the stops to control the air flow into her organ’” (Rayo 2011). It is not too difficult to
envisage a grab-bag for the word ‘novel’, which would include information about the
stories, plots and characters that authors imagine, the written (or virtual) texts they may
produce as a result, the publication, printing, selling and distribution processes, and the
resulting physical copies. On different occasions of use, different selections are made from
the grab-bag, in accordance with ‘common sense and sensitivity to context’, which I take
to be a (somewhat cavalier) reference to the cognitive interpretive processes that a
pragmatic theory seeks to explain.
10
The instantiation of ‘meaning eliminativism’ that Recanati (2004: 146-151) sets out is rather different
from the one discussed here. On the account he outlines, what a word form brings to the interpretation
process is a stored collection of its previous uses and its interpretations in context.
 17

An approach along these wholly pragmatic lines would put an end to the need for
an array of semantic rules, even supposing they are formulable, and the futile attempts to
decide which of the various senses of polysemous words to take as the basic one. Of
course, it remains to be spelled out in detail exactly how the grab-bag selection process
works, particularly how it can result in a concept with a truth-conditional content, but the
basic intuition seems to be very much in keeping with the ad hoc concept construction
process in relevance theory, as outlined in Section 2.

5. Conclusion

The goal of this paper was modest: to present a range of reasons for taking seriously the
idea that words (or lexical forms) may not encode concepts or map directly to contentful
entities, but rather come with meaning-relevant components that are different in kind from
semantic values, that are intrinsically underspecified with regard to content, where a
content is what is expressed/communicated by an individual’s use of a word and so is only
determinable on an occasion of use. Thus, this hypothesis is only worth exploring when
coupled with a well-developed pragmatic theory that seems capable of providing a
detailed account of how the concepts a speaker intends to express can be recovered by her
addressee on the basis of such underspecified meanings or encyclopaedic information.
Abandoning a conceptual lexical semantics raises a host of new questions. First,
there is the issue of maintaining a distinction between genuinely indexical words and these
cases of word meaning modulation; on a non-conceptual construal of word type meaning,
the pragmatic process of finding an appropriate semantic value is no longer optional, so
cannot be distinguished on those grounds from the obligatory process of indexical
saturation. Second, the approach has consequences for (relevance-theoretic) pragmatics, in
that we can no longer think in terms of the narrowing or broadening of denotations (or of
concept adjustment) as there is no linguistically-specified denotation to narrow or broaden
(and no concept to adjust). All concepts occurring in communicated thoughts
(explicatures) are pragmatically inferred and merely constrained by an encoded lexical
schema/template or an array of activated encyclopaedic information (a grab-bag). In that
sense, all concepts expressed or communicated are ‘ad hoc’. Third, there is a robust
intuition that many words have a ‘literal meaning’, that there is a particular concept (or
concepts) which, among the others that the word may be used to express, is somehow
 18

privileged or basic. Whether and, if so, how this intuition is to be respected on any of
these non-conceptual approaches to word type meaning needs to be addressed. Finally, the
most pressing question is how to account for the move from a non-conceptual, non-
semantic entity to a conceptual, contentful one. The project ahead, as I see it, is to
investigate each of these questions, in conjunction with various instantiations of the wrong
format or eliminativist views of word meaning, within the explanatory pragmatic account
provided by the relevance-theoretic framework.

References:

Asher, Nicholas. 2011. Lexical meaning in context: A web of words. Cambridge:

Cambridge University Press.
Atlas, Jay. 1989. Philosophy without ambiguity. Oxford: Clarendon Press.
Borer, Hagit. 2005. In name only: Structuring sense, vol. 1. Oxford: Oxford University
Press.
Bosch, Peter. 2007. Productivity, polysemy, and predicate indexicality. In Balder ten Cate
& Henk Zeevat (eds.), Proceedings of the sixth international Tbilisi symposium on
language, logic & computation, 58-71. Berlin: Springer.
Carston, Robyn. 2002. Thoughts and utterances: The pragmatics of explicit
communication. Oxford: Blackwell.
Carston, Robyn. 2012. Relevance theory. In Gillian Russell & Delia Graff Fara (eds.),
Routledge companion to the philosophy of language, 163-176. London: Routledge.
Carston, Robyn and Catherine Wearing. 2011. Metaphor, hyperbole and simile: A
pragmatic approach. Language and Cognition 3. 283-312.
Cohen, Jonathan. 1993. The semantics of metaphor. In Andrew Ortony (ed.), Metaphor
and thought, 2nd edn., 64-77. Cambridge: Cambridge University Press.
Fodor, Jerry A. 1998. Concepts. Where cognitive science went wrong. Oxford: Clarendon
Press.
Fodor, Jerry A. 2001. Language, thought and compositionality. Mind and Language 16. 1-
15.
Fodor, Jerry A. 2008. LOT 2: The Language of Thought Revisited. Oxford: Clarendon
Press.
 19

Fodor, Jerry A. and Ernest Lepore. 1998. The emptiness of the lexicon: Reflections on
Pustejovsky. Linguistic Inquiry 29(2). 269-288.
Glanzberg, Michael. 2011. Meaning, concepts and the lexicon. Croatian Journal of
Philosophy 11. 3-31.
Kempson, Ruth, Eleni Gregoromichelaki & Christine Howes. 2011. Introduction. In Ruth
Kempson et al. (eds.) The Dynamics of Lexical Interfaces. 1-20. Stanford: CSLI
Publications.
Lossius Falkum, Ingrid. 2011. The semantics and pragmatics of polysemy. PhD
dissertation, University College London.
Moravcsik, Julius. 1994. Is snow white? In Paul Humphreys (ed.), Patrick Suppes:
Scientific Philosopher. 71-85. The Netherlands: Kluwer Academic Publishers.
Murphy, Gregory. 2002. The big book of concepts. Cambridge Mass.: MIT Press.
Nunberg, Geoffrey. 1979. The non-uniqueness of semantic solutions: polysemy.
Linguistics and Philosophy 3. 143-184.
Pustejovsky, James. 1995. The generative lexicon. Cambridge Mass.: MIT Press.
Rayo, Agustín. forthcoming. A plea for semantic localism. Noûs
Recanati, François. 2004. Literal meaning. Cambridge: Cambridge University Press.
Ruhl, Charles. 1989. On monosemy: A study in linguistic semantics. Albany: State
University of New York Press.
Sperber, Dan & Deirdre Wilson. 1995. Relevance: Communication and cognition. Oxford:
Blackwell.
Sperber, Dan & Deirdre Wilson. 1998. The mapping between the mental and the public
lexicon. In Peter Carruthers & Jill Boucher (eds.), Language and thought, 184-200.
Cambridge: Cambridge University Press.
Textor, Mark & Nicholas Allott. forthcoming. Lexical pragmatic adjustment and the
nature of ad hoc concepts. International Review of Pragmatics.
Wilson, Deirdre & Robyn Carston. 2007. A unitary approach to lexical pragmatics. In
Noel Burton-Roberts (ed.), Pragmatics, 230-260. Basingstoke: Palgrave.